Application of Finite Mixture of Regression Model with Varying Mixing Probabilities to Estimation of Urban Arterial Travel Times

Abstract

Travel time along an urban arterial is greatly affected by traffic signals. Most studies on urban travel time use statistical models to obtain the distribution directly without incorporating the effects of traffic signal timing. In this study, a finite mixture of regression model with varying mixing probabilities (weights) was proposed to gain a better understanding of urban travel time distribution through consideration of signal timing. Standard finite mixture models with constant mixing probabilities have a limited ability to adapt to underlying random structural changes for observed travel times. The model developed in this study can capture such dynamics by ( a) modeling the mixing probabilities as a function of the explanatory variables associated with signal timing and ( b) establishing a linear regression between the mean of each component and signal timing. The finite mixture of regression model was applied to the travel time data collected by the automatic vehicle identification system on one urban arterial with the Sydney coordinated adaptive traffic system (SCATS). The results demonstrate that the varying mixing probabilities can be used to classify the samples of travel time, and the mean values of components can capture the effects of signal timing. By comparing various types of mixture models, the proposed approach not only has a better statistical fitting performance but also provides useful information about travel time features.